The satisfaction of client needs is the goal of most of the industrial systems, which can be achieved by appropriate life-cycle management. When it concerns complex real-world systems, the difficulty of managing their life-cycle increases with increasing of the links and interactions between the system components and between the system and its environment. Therefore, the need of addressing a complete and realistic maintenance planning approach to face these difficulties is crucial. This article presents a methodology for maintenance optimization of complex systems using Bayesian networks. In this methodology, the objective function, which aims at maximizing the system benefit, allows conciliating between two contradictory objectives: reducing the maintenance costs and reaching an availability target fixed according to the customer demand. The Bayesian networks are used to take into account the system interactions, while the maintenance policy, which is based on the imperfect preventive maintenance and considers several efficiency levels, is used to build a realistic maintenance planning model. An application to a water supply system is included to illustrate the benefit and the effectiveness of the proposed approach.